CN114169216A - Multi-agent heterogeneous target cooperative coverage method based on self-adaptive partitioning - Google Patents
Multi-agent heterogeneous target cooperative coverage method based on self-adaptive partitioning Download PDFInfo
- Publication number
- CN114169216A CN114169216A CN202111229814.9A CN202111229814A CN114169216A CN 114169216 A CN114169216 A CN 114169216A CN 202111229814 A CN202111229814 A CN 202111229814A CN 114169216 A CN114169216 A CN 114169216A
- Authority
- CN
- China
- Prior art keywords
- target
- agent
- target point
- point
- angle
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000000638 solvent extraction Methods 0.000 title abstract description 12
- 238000005192 partition Methods 0.000 claims abstract description 29
- 230000002068 genetic effect Effects 0.000 claims abstract description 25
- 238000001514 detection method Methods 0.000 claims abstract description 16
- 230000008439 repair process Effects 0.000 claims abstract description 14
- 238000003064 k means clustering Methods 0.000 claims abstract description 10
- 108090000623 proteins and genes Proteins 0.000 claims description 41
- 238000005457 optimization Methods 0.000 claims description 17
- 210000000349 chromosome Anatomy 0.000 claims description 12
- 230000035772 mutation Effects 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 12
- 230000003044 adaptive effect Effects 0.000 claims description 8
- 239000002245 particle Substances 0.000 claims description 5
- 238000013316 zoning Methods 0.000 claims description 5
- 238000009826 distribution Methods 0.000 claims description 4
- 241000507649 Kerria japonica Species 0.000 claims description 3
- 230000001131 transforming effect Effects 0.000 claims description 3
- 101150100366 end gene Proteins 0.000 claims description 2
- 230000002123 temporal effect Effects 0.000 claims 1
- 230000007246 mechanism Effects 0.000 abstract description 4
- 230000008901 benefit Effects 0.000 description 7
- 238000010586 diagram Methods 0.000 description 7
- 238000012217 deletion Methods 0.000 description 4
- 230000037430 deletion Effects 0.000 description 4
- 230000014509 gene expression Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000013439 planning Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000010429 evolutionary process Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012994 industrial processing Methods 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000000241 respiratory effect Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 241000894007 species Species 0.000 description 1
- 108700026220 vif Genes Proteins 0.000 description 1
- 230000004580 weight loss Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Biophysics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Software Systems (AREA)
- Medical Informatics (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Probability & Statistics with Applications (AREA)
- Physiology (AREA)
- Genetics & Genomics (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Feedback Control In General (AREA)
Abstract
The invention provides a multi-agent heterogeneous target cooperative coverage method based on self-adaptive partitioning, which can realize multi-agent multi-target point coverage under the condition of resource limitation and balance the time loss of each agent. The invention provides an improved k-means clustering algorithm based on a feedback control mechanism to partition a target point and adaptively adjust the program punctuation partition result, thereby further balancing the time loss of each intelligent agent for executing tasks. In the stage of optimizing the traversal sequence of each agent on the obtained target points, the genetic algorithm is improved based on the problem of multi-agent multi-target-point coverage under the condition of considering insufficient resources, so that the detection yield is maximized as far as possible within limited time. Considering that the movement time of each intelligent agent is limited, a heuristic repair rule path repair method is designed, so that each intelligent agent is ensured to return to the base station within a preset time range.
Description
Technical Field
The invention relates to the technical field of multi-agent multitask collaboration and intelligent optimization, in particular to a multi-agent heterogeneous target collaborative coverage method based on self-adaptive zoning.
Background
With the rapid development of the robot technology and the sensor technology, the multi-robot system is widely applied to various industries, such as environmental quality monitoring, industrial production and processing, urban logistics transportation, disaster area emergency rescue, military operation reconnaissance and the like. Compared with a single robot system, the multi-robot system has higher robustness, intelligence and cooperativity, and can greatly improve the task execution efficiency and shorten the task execution time through effective coordination and division of labor among multiple robots.
In order to realize effective coordination and division of labor among the agents, the problem of effective coordination and division of labor among the agents is converted into a combined optimization problem of multi-agent multi-target cooperative coverage, and the problem of multi-agent multi-target cooperative coverage mainly refers to the following steps: according to the method, a plurality of target points and a plurality of agents to be executed in a given task environment, how to reasonably divide the task load of the agents and effectively schedule the target sequence covered by each agent enables each agent to maximize the task execution profit or minimize the overall system loss and the like under the condition of meeting multiple constraints such as self constraint, target task demand constraint and the like. The description of the multi-agent multi-target cooperative coverage problem can be abstracted into a multi-agent cooperative path planning problem. For example, the Vehicle Routing Problem (VRP) refers to how to achieve effective allocation between vehicles and customers and reasonable arrangement of vehicles to commodity transportation sequence (customer arrival sequence) under the condition that multiple constraints are met for a given number of vehicles, so as to achieve multiple purposes of minimum transportation cost, maximum system benefit, highest customer satisfaction and the like.
However, in the existing multi-agent multi-target point cooperative coverage method, there are few multi-agent multi-target point coverage problems under the condition of resource shortage, such as the vehicle routing problem (CVRP), Dynamic Vehicle Routing Problem (DVRP), batch cargo Vehicle Routing Problem (VRPSD) and the like, which are presented as the variants of VRP problem in the existing research, and the planning targets of the above problems are mainly focused on minimizing the number of resources, minimizing the transportation loss or maximizing the customer satisfaction, there is few problems that the vehicle resources are limited in the multi-vehicle routing problem, especially in the disaster area rescue scene, the multi-agent needs to detect the areas with different importance, and considering the limitation of energy of sensing devices such as vehicles, each vehicle needs to return to the base station for energy replenishment before the energy is exhausted (special vehicle transportation is needed if the energy of the vehicle is exhausted during the task execution), and further to maximize the detection yield as much as possible within a limited time. In addition, according to the conventional multi-agent heterogeneous target collaborative coverage method, the task point clustering result based on the k-means algorithm is greatly influenced by the initial central point, and if the initial central point is not properly selected, an effective clustering result is difficult to obtain, so that the optimality of the target coverage scheme of each agent is further influenced.
Disclosure of Invention
In view of this, the invention provides a multi-agent heterogeneous target cooperative coverage method based on adaptive partitioning, which can realize multi-agent multi-target point coverage under the condition of resource limitation and balance the time loss of each agent.
In order to achieve the purpose, the technical scheme of the invention is as follows:
a multi-agent heterogeneous target collaborative coverage method based on adaptive zoning comprises the following steps:
the method comprises the steps that firstly, an optimization problem model for optimizing a target point distribution scheme and a target point traversal sequence of each intelligent agent to maximize detection yield is established for a multi-intelligent-agent heterogeneous target with heterogeneous detection weight values and coverage time; inputting a problem parameter and a genetic algorithm parameter into the model;
secondly, adopting a k-means clustering algorithm to output based on the angular density of the targetThe angle value of each clustering center is obtained, and the target is partitioned; wherein the angular density of the target node i is as large as
Wherein theta isiAnd thetajRespectively representing the angles between the connecting lines of the target node I, the target node j and the base station and a horizontal line, wherein delta theta is a set angle threshold value, and I represents a counting function; i is 1,2,3 … m, j is 1,2,3 … m, m is the target point, and the total number i is not equal to j;
thirdly, optimizing the target point traversal sequence in each partition based on an improved genetic algorithm to obtain the optimal traversal sequence of the target points in each partition; the improved genetic algorithm comprises the following steps:
step 1: initializing a population by adopting a hybrid heuristic rule and a Kerries-based heuristic algorithm;
step 2: performing cross operation in an individual cross mode combining the global guidance idea of the particle swarm algorithm;
and step 3: performing mutation operation;
and 4, step 4: calculating the fitness value of individuals in the population;
and 5: updating the population by adopting an elite strategy, which specifically comprises the following steps: sorting all the individuals from big to small based on the objective function value, and selecting the top NpThe good individual enters the next iteration, where NpIs a set value;
step 6: judging whether the end condition of the improved genetic algorithm is reached, if so, continuing the step 7, and if not, returning to the step 2;
and 7: outputting the optimal traversal sequence of the target points in the partitions;
fourthly, calculating the difference value between the execution time of each agent task and the mean value of the execution times of all the agent tasks, and adjusting the corresponding feedback parameters of the zones with the difference values not being 0 according to the difference values;
fifthly, judging whether the current difference value meets the set finishing condition, if not, returning to the second step according to the adjusted feedback parameter, and re-clustering the target point; if yes, continuing the sixth step;
deleting target points which do not meet the time constraint under the obtained optimal traversal sequence, and repairing the optimal traversal sequence;
and seventhly, outputting the traversal sequence of the target points of the repaired intelligent agents.
In the second step, in the k-means clustering algorithm, the method for acquiring the initialization center point comprises the following steps: selecting the angle of the target with the maximum angular density value as a central point, then removing the selected target point and all points in the angle neighborhood of the selected target point, calculating the angular density of the rest target points, selecting the angle of the target with the maximum angular density value as the central point again, and repeating the process until the number of the central points is the same as that of the intelligent agents;
the method for adjusting the clustering center line comprises the following steps: firstly, obtaining the number of a central line closest to each target point according to a formula (1.12); then, calculating the angle mean value of target points belonging to various types according to the formula (1.13), and updating the angle of the clustering center line; finally, calculating the variation of the central line before and after updating according to the formula (1.14), and if the variation of the central line before and after updating is less than a preset threshold value deltaangleEnding the loop and outputting the angle value of each clustering center, wherein:
in the formula (1.12), σiIndicating the angle value, mu, between the target point i and the base station connecting line and horizontal linejRepresents the angle of the centerline j; in formula (1.13) < mu'jRepresents the angle of the updated centerline j; in the formula (1.14), K represents the number of cluster centers.
In the third step, the specific implementation manner of step 2 of the improved genetic algorithm is as follows:
firstly, a global edge relation table is constructed, and the percentage of the next position of each target point in the existing population, which is connected with target points of different types, is counted in the table; and then crossing the individuals meeting the requirements, randomly selecting chromosome segments on the parent individuals, transforming the parent chromosomes based on the edge relation table, and storing the generated child chromosomes in the population.
The crossing operation comprises four crossing modes, and the specific process is as follows:
Cross-I operation: combining the residual genes and the global edge relation table to replace the genes in gs;
Cross-II operation: keeping the gene sequence in gs unchanged, and replacing the residual genes by combining the global edge relation table;
Cross-III operation: keeping the gene sequence in gs unchanged, using the gene sequence as an initial gene segment of a filial generation, and sequentially generating subsequent genes by combining a global edge relation table;
Cross-IV operation: keeping the gene sequence in gs unchanged, using the sequence as the end gene segment of the filial generation, and combining the rest genes with the global edge relation table to generate in sequence.
In the third step, in the step 3 of the improved genetic algorithm, mutation operation is performed on individuals satisfying the condition by adopting a gene point exchange mode.
And repairing the target traversal sequence which does not meet the time constraint by using a heuristic rule path repairing method.
The heuristic rules comprise a heuristic repair rule based on time weight ratio and a heuristic repair rule based on greedy distance.
Has the advantages that:
the invention provides an improved k-means clustering algorithm based on a feedback control mechanism to partition a target point, balance the time loss of each intelligent agent for executing tasks, dynamically adjust feedback parameters for time information completely covered by the allocated target point by combining each intelligent agent, and adaptively adjust program punctuation partition results, thereby further balancing the time loss of each intelligent agent for executing tasks. In the stage of optimizing the traversal sequence of each agent on the distributed target points, the genetic algorithm is improved based on the problem of multi-agent multi-target-point coverage under the condition of considering insufficient resources, and the optimization problem model is designed for effectively considering the diversity of the population and the convergence of the algorithm and for the population initialization and cross operator. In the aspect of population initialization, a hybrid heuristic rule and an initialization rule based on a Christofises heuristic (Christofises Heuristic) algorithm are designed. A new crossover operator is provided in the aspect of individual crossover, so that the detection benefit is maximized as far as possible in a limited time. Considering that the movement time of each intelligent agent is limited, a heuristic repair rule path repair method is designed, so that each intelligent agent is ensured to return to the base station within a preset time range.
Drawings
FIG. 1 is a schematic diagram of a multi-agent multi-target collaborative coverage problem according to an embodiment of the present invention;
FIG. 2 is a frame diagram of a multi-agent heterogeneous target cooperative coverage method based on adaptive zoning according to an embodiment of the present invention;
FIG. 3 is a flow chart of a multi-agent heterogeneous target cooperative coverage algorithm based on adaptive zoning according to an embodiment of the present invention;
FIG. 4 is a schematic illustration of the angular density of target points in an embodiment of the present invention;
FIG. 5 is a flow chart of the k-means algorithm center point initialization in an embodiment of the present invention;
FIG. 6 is a flow chart of a target point traversal order optimization algorithm based on an improved genetic algorithm in an embodiment of the present invention;
FIG. 7 is a schematic diagram of a hybrid heuristic initialization rule in an embodiment of the invention;
FIG. 8 is a flow chart of the Christofides respiratory algorithm in an embodiment of the present invention;
FIG. 9 is a schematic diagram illustrating a heuristic initialization rule based on a Christofises juristic algorithm in an embodiment of the present invention;
FIG. 10 is a diagram illustrating the generation of an edge relationship table according to an embodiment of the present invention;
FIG. 11 is a graph showing the results of four individual crossover chromosomes based on the edge relation table in the example of the present invention;
FIG. 12 is a flow chart of an individual intersection algorithm based on an edge relation table according to an embodiment of the present invention;
FIG. 13 is a diagram illustrating mutation operators in accordance with an embodiment of the present invention.
Detailed Description
The invention is described in detail below by way of example with reference to the accompanying drawings.
The invention discloses a multi-agent heterogeneous target collaborative coverage method based on self-adaptive partitioning, which is used for developing research aiming at the problem of multi-agent heterogeneous target collaborative coverage under the condition of limited return time, aiming at a plurality of types of target points with heterogeneous detection weight values and coverage time, and optimizing a target point distribution scheme and a target point traversal sequence of each agent so as to maximize detection benefits. The adaptive partitioning refers to further adjusting the partitioning condition of the target point according to related data (such as time required by each agent to traverse the target) obtained by the target point partitioning last time, and the purpose of optimization is achieved through multiple adjustments. And finally, deleting the targets on each path by adopting two heuristic rules so as to meet the task execution time constraint of the intelligent agent.
The invention introduces the concept of angular density and initializes the central point of the clustering algorithm. Designing an improved k-means algorithm based on feedback control, combining feedback data of the result of optimization of the covering sequence of the target point in each partition every time, adjusting control factors of each clustering center point, and further performing self-adaptive dynamic adjustment on the clustering result; the optimization result of the covering sequence of the target point in each partition is the time shortest path generated in each partition, the optimization result of the covering sequence of the target point in each partition is obtained according to an improved genetic algorithm, two kinds of population initialization rules are adopted for initialization in the improved genetic algorithm, and meanwhile, an individual intersection mode combining the global guidance idea of a Particle Swarm Optimization (PSO) is adopted for intersection operation, so that the detection yield is maximized as far as possible within limited time.
In the embodiment of the cooperative coverage method of the multi-agent heterogeneous target based on the adaptive partitioning, which is provided for the problem of cooperative coverage of the multi-agent heterogeneous target point in the disaster area rescue environment, for abstract description of the problem, refer to fig. 1, where each closed loop represents a traversal order of task points of each agent. Compared with the prior art, the algorithm has the advantages in the aspects of algorithm operation efficiency and target point weight coverage income.
The framework schematic diagram of the multi-agent heterogeneous target collaborative coverage method based on the adaptive partitioning in the embodiment is shown in fig. 2, and includes three aspects: firstly, a target with a given position is subjected to initialization partitioning, and the target point traversal sequence in each partition is optimized based on an improved genetic algorithm. And then, further adjusting the partition scheme of the target point by adopting a feedback control mechanism in combination with the task execution time of the agents in each partition until the corresponding index of the partition scheme reaches the algorithm ending condition. And finally, deleting the task load executed by each agent by combining the time constraint of the agent, wherein the deletion is specifically represented as the deletion of a target point on the task execution path of each agent.
The specific process of the method of the present invention is shown in FIG. 3. The method comprises the following steps:
the method comprises the steps that firstly, an optimization problem model for optimizing a target point distribution scheme and a target point traversal sequence of each intelligent agent to maximize detection benefits is established for multi-type target points with different detection weight values and different coverage times; inputting problem parameters and genetic algorithm parameters into the model, wherein the problem parameters comprise target point positions, target point weights, target execution time, the number of intelligent agents, base station positions, time thresholds of tasks executed by the intelligent agents and the like, and the genetic algorithm parameters comprise population scale, maximum iteration times, cross probability, variation probability, angle neighborhood values and the like;
given n agents, m target points to be covered and a base station, the above multi-agent heterogeneous target point cooperative coverage problem may be modeled as G ═ a, E, where a ═ aT∪A0Represents a set of nodes, AT1,2, …, m represents a set of target points, a0Indicating the base station (starting position of each agent). E denotes an arc connecting points, each eyeThe distances between the punctuations, the target point and the base station are expressed by Euclidean distances. The task execution time at each target point is heterogeneous, the detection values (importance) of the target points are different, and the multi-agent needs to maximize the detection yield under the condition that the task execution time is limited, namely, the overall weight coverage value is maximized, and the mathematical expression of the overall weight coverage value is shown as a formula (1.1).
In the formula (I), the compound is shown in the specification,
a decision variable representing the agent k going from the current target i to the next target j, n represents the total number of agents, m represents the total number of targets, "0" represents the base station number, ωjRepresenting the detection yield value of target point j.
The above optimization problem is affected by various constraints, as detailed in the problems shown in formulas (1.2) to (1.11).
The formula (1.2) is the flow balance constraint of each target point, namely, if the inflow (arrival) of the agent exists at each target point, the outflow (departure) of the agent exists; formulas (1.3) - (1.4) show that each agent can only detect one target area at a time and does not go back and forth; equation (1.5) represents the agent quantity constraint, i.e. the upper limit of the target point sequence quantity; the formula (1.6) represents that the working time of each agent does not exceed the preset time constraint; equation (1.7) represents the sub-loop cancellation constraint; formulas (1.8) - (1.9) indicate that each agent needs to start from the base station and return to the base station; equation (1.10) represents a decision variable type; formula (1.11) represents μiThe value range of (a).
Secondly, partitioning the target with the given position by adopting a k-means clustering algorithm based on the angular density, wherein the specific analysis is as follows:
the traditional k-means clustering algorithm aggregates points with similar characteristics by dynamically adjusting the position of a clustering central point, namely, blocks data points, is greatly influenced by an initial central point, and an effective clustering result cannot be obtained if initial value selection is wrong. In consideration of the problem to be solved by the invention, each agent is sent from the base station, and the agent returns to the base station after traversing the divided target point in an optimal way, namely, a traveler problem is completed, so that a simple Euclidean distance-based point clustering way is not applicable to the problem.
The invention improves the traditional k-means clustering algorithm and provides a k-means clustering algorithm based on angular density, wherein the angular density of a node i is defined as
Wherein theta isiAnd thetajRespectively representing the angles between a point I and a point j and between a base station connecting line and a horizontal line, wherein delta theta is a set angle threshold value, I represents a counting function, and counting is carried out when the condition is met. As shown in FIG. 4, the number of points in the three sector regions respectively represents the angular density value of the point where the arrow is located in each sub-region, where ρI>ρII>ρIII。
The flow of the initialization center point acquisition method is shown in figure 5, and according to the position of a target point, the position of a base station, the number of agents (determining the number of partition center points) and delta theta; and sequentially calculating the angular density of each target point, selecting the angle of the target with the maximum angular density value as a central point (the central point is a line because of angle clustering) each time, then removing all points in the selected target point and the angle neighborhood thereof, and calculating the angular density of the remaining target points until the number of the central points is the same as that of the intelligent agents. Wherein, in the adjusting process of the clustering center line: firstly, obtaining the number of a central line closest to each target point according to a formula (1.12); then, calculating the angle mean value of target points belonging to various types according to the formula (1.13), and updating the angle of the clustering center line; finally, the amount of change in the center line before and after updating is calculated according to equation (1.14). If the two-time change value is less than the set threshold value deltaangleAnd ending the circulation and outputting the angle value of each clustering center. Wherein:
in the formula (1.12), σiIndicating the angle value, mu, between the target point i and the base station connecting line and horizontal linejRepresents the angle of the centerline j; in formula (1.13) < mu'jRepresents the angle of the updated centerline j; in the formula (1.14), K represents the number of cluster centers.
Thirdly, optimizing the target point traversal sequence in each partition based on an improved genetic algorithm according to problem parameters (the target point position and the base station position) and genetic algorithm parameters (the population scale, the maximum iteration time, the cross probability and the variation probability) to obtain the optimal traversal sequence of the target point in each partition; the specific algorithm flow is shown in fig. 6, and comprises the following steps:
step 1: initializing a population by adopting a hybrid heuristic rule and a Kerries-based heuristic algorithm;
step 2: performing cross operation in an individual cross mode combining the global guidance idea of the particle swarm algorithm;
and step 3: performing mutation operation;
and 4, step 4: calculating the fitness value of individuals in the population;
and 5: updating the population by adopting an elite strategy;
step 6: judging whether an algorithm ending condition is met, if so, continuing to step 7, and if not, returning to step 2;
and 7: and outputting the optimal traversal sequence of the target points in the partitions.
The expression of the solution in the method mainly refers to the traversal sequence of the targets obtained by each agent, each target has a self number, the solution of the problem is expressed by adopting an integer coding mode, and the population size is defined as Np.
The step 1 specifically comprises the following steps:
in the evolutionary algorithm, the initial solution plays a crucial role in the evolution process of the whole population, and the excellent initial solution can guide the population to search towards a more effective solution space direction, so that the convergence of the algorithm is accelerated. Therefore, the invention adopts two kinds of population initialization rules in combination with the problem characteristics: (1) mixing heuristic rules; (2) and (3) heuristic rules based on Christofides hemristic algorithm.
(1) The mixed heuristic rule specifically comprises the following steps:
considering that the optimization goal of the target point traversal order is to minimize the time loss of traversing all the points as much as possible, when generating each path scheme, the point with the minimum time loss from the current point is selected with a certain probability, and the algorithm flow is shown in fig. 7. The algorithm inputs are the position of the target point, the target execution time, the base station position, the threshold β (whether the decision is to select in a random or greedy manner), and the population size. The output of the algorithm is the population initialization result.
The distance between two target points in the invention is expressed by Euclidean distance, as shown in formula (1.16).
In the formula (x)i,yi) Coordinates representing object i, (x)j,yj) Representing the coordinates of object j.
The time required for the agent to move from the current point i to the next task point j is expressed mathematically as in equation (1.17). Task execution time at due to the existence of target pointsjThe agents need to leave the target point after meeting the task execution time and continue the subsequent target, so the time loss of each agent in the whole traversal sequence is shown as the formula (1.18).
(2) The heuristic rule based on the CrisStokes algorithm is specifically as follows:
the christofield algorithm (Christofides algorithm) is an approximation algorithm of the travelling quotient problem in the metric space (i.e. distance symmetry and satisfying the triangle inequality), and can ensure that the relative optimal hamiltonian loop length has an approximate ratio of 3/2, and the algorithm flow is shown in fig. 8. The algorithm is used for generating a first individual in a population initialization process, and corresponding operation is carried out by taking the first individual as a reference individual, so that the whole population is initialized, and the method specifically comprises the following steps:
in this embodiment, a clistofides algorithm-based heuristic rule initializes a population, which is shown in detail in fig. 9. The input of the algorithm is mainly the position of a target point, the target execution time, the position of a base station and the population scale. The output of the algorithm is the population initialization result. First, Christofide algorithms is adopted to carry out initialization assignment on a first individual to obtain a better solution pop relative to the approximation degree of the optimal solution 3/2. The remaining individuals are then initialized in conjunction with a stochastic algorithm. In order to keep better-solved good genes, a gene segment on the pop of an individual is randomly selected as an initial gene segment pop 'of a new individual, and then the rest individuals on the pop are sequentially arranged behind the pop' to form the new individual. The population initialization mode simultaneously considers the excellent gene retention and the diversity exploration of the better solution.
The step 2 specifically comprises the following steps:
the crossover operator is the most main genetic operator in the genetic algorithm, and the diversity of the population and the convergence of the algorithm can be better balanced through individual crossover, thereby playing an important role in improving the population search performance. In consideration of the traditional crossing modes such as uniform crossing and the like, the connection relation between target points on individual gene segments is not considered when new individuals are generated, so that the excellent target point ordering relation in a parent is easily ignored. The invention uses the thought that the individual search in the particle swarm algorithm is simultaneously guided by the cooperation of the individual local information and the group global information, and records the connection relation between each task point in the parent group by constructing the global edge relation table, thereby guiding the search behavior of the offspring individuals in the problem space.
Fig. 10 shows a process of constructing the edge relation table, where the target point number "2" and its neighbor point are marked as blue in the left table, and taking the first individual as an example, if the neighbors of the "target point 2" are "target point 4" and "target point 1", then the corresponding numbers are marked in the "global edge relation table", which is detailed in the portion marked as blue in the right table, and the neighbors counted as "target point 2" are: 4,1,7,8,10,3,12.
The present invention contemplates four interleaving modes, as shown in fig. 11. The red dotted line box is marked with randomly intercepted gene segments (called gs in the following description), and the brown dotted line box is marked with 4 corresponding intersection modes, and the specific process is as follows:
Cross-I operation: and combining the residual genes and the global edge relation table to replace the genes in gs.
Cross-II operation: and (5) keeping the gene order in gs unchanged, and replacing the residual genes by combining the global edge relation table.
Cross-III operation: keeping the gene sequence in gs unchanged, using the sequence as an initial gene segment of a filial generation, and sequentially generating subsequent genes by combining a global edge relation table.
Cross-IV operation: the sequence of the genes in gs is kept unchanged and is used as the terminal gene segment of the filial generation, and the other genes are combined with the global edge relation table to generate in turn.
The flow of the individual intersection implementation based on the global edge relation table is shown in fig. 12. The inputs to this algorithm are mainly population chromosomes and crossover probabilities. The output of the algorithm is the population after crossing. Firstly, a global edge relation table is constructed, and the percentage of the next position of each target point in the existing population, which is connected with different types of target points, is counted in the table. And then crossing the individuals meeting the requirements, randomly selecting chromosome segments on the parent individuals, transforming the parent chromosomes based on the edge relation table, and storing the generated child chromosomes in the population. Through the crossing process, partial gene segments of the parent individuals are reserved in the filial individuals, and meanwhile, excellent gene combinations are reserved under the influence of a population 'global edge relation table'. The design of the individual crossover operator starts from problem knowledge, and not only inherits excellent genes, but also maintains the diversity of individuals.
The mutation operator performs mutation on individuals by simulating mutation links in the biogenetic and evolutionary processes, and although the probability of the mutation is low, the mutation operator is a crucial way for generating new species. Similar to the crossover operator, in consideration of the specificity of the target point sequence optimization problem, in order to further improve the optimization effect, in the genetic algorithm of the present invention, mutation operation is performed on individuals satisfying the conditions by using a gene point exchange method, as shown in fig. 13, and yellow and blue gene positions are exchanged to obtain a new chromosome.
For the individual objective function, when optimizing the traversal order of the target points divided by the agent, the target points need to be completely traversed, so the advantages and disadvantages of the traversal order are irrelevant to the execution time of the target points (since the sum of the target execution times inside each partition is determined after the partition scheme is determined), and only depending on the traversal order, namely the moving order of the agent between the target points, the optimization objective is to minimize the total time length of the agent traversing all the divided targets, as detailed in formula (1.18).
In each iteration process, due to the introduction of a crossover operator and a mutation operator, the population size is larger than a set value NpTherefore, individuals need to be screened. The population is updated by adopting an elite selection strategy, and the method is concretely as follows. Sorting all the individuals from big to small based on the objective function value, and selecting the top NpAnd entering the next iteration for one good individual.
Fourthly, calculating the difference value between the execution time of each agent task and the mean value of the execution times of all the agent tasks, and adjusting the corresponding feedback parameters of the zones with the difference values not being 0 according to the difference values;
since the clustering target of the invention is to minimize the execution time of various (partitioned) agents traversing the partitioned target (specifically including the task execution time of the target area itself and the agent movement time, see the formula (1.18)), the target point segmentation scheme obtained only by the traditional clustering method is no longer applicable. The clustering algorithm based on the feedback control mechanism is provided, and the time loss of the traversal sequence of the optimal target point of the intelligent agent at various types (partitions) is balanced as much as possible by dynamically adjusting various feedback parameters. The specific adjustment scheme is as follows:
let t be the time loss corresponding to class (partition) iiTime-loss mean of all classes (partitions)
Is shown as
From this, the difference between the time loss and the mean loss at various classes (partitions) can be calculated
For Δ tiThe class (partition) not equal to 0 adopts the formula (1.15) to correspond to the feedback parameter etaiAnd (6) carrying out adjustment.
ηi＝ηi+κ·Δti (1.15)
Where κ is a control constant.
Fifthly, judging whether the current difference value meets the set finishing condition, if not, returning to the second step according to the adjusted feedback parameter, and re-clustering the target point; if yes, continuing the sixth step;
sixthly, repairing the traversal sequence of each agent based on time constraint;
considering that each agent has task execution time constraint and the overall goal is to maximize the detection yield as much as possible, the 'poor' point in the sequence is deleted based on problem knowledge when the target point deletion is performed. Therefore, before the target point traversal sequence of each agent is output, the traversal sequence of each agent is repaired based on the time constraint, the target traversal sequence which does not meet the time constraint is repaired by using the heuristic rule, and then the target point traversal sequence is output. Specifically, two heuristic repair rules are designed, namely a repair rule based on time weight ratio and a repair rule based on greedy time.
The heuristic repair rule based on the time weight ratio specifically comprises the following steps: because the invention aims to improve the weight coverage yield of the target point as much as possible under the condition of the execution time constraint of each intelligent agent task, the time difference caused by deleting each point needs to be calculated in the deleting process, and the expressions (1.19) - (1.20) are specifically described by combining the preset weight loss of the target point.
Δtj＝Tc-Tcwithoutj (1.19)
Equation (1.19) represents the time difference between the sequence of target point execution by the agent before and after target point deletion; the formula (1.20) is the selection principle of deleting the target point.
The heuristic repair rule based on the distance greedy is specifically as follows: because the task execution time of the intelligent agent exceeds the set time threshold, in order to effectively reduce the task execution time, the Euclidean distance between each target point and the base station needs to be calculated, and the point p with the largest distance is deleted from the target point sequence every time*Until the task execution time is less than the time threshold.
In the formula d0jRepresenting the euclidean distance of the target point j to the base station.
And seventhly, outputting the target point traversal sequence of each repaired intelligent agent.
In summary, the above description is only a preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (7)
1. A multi-agent heterogeneous target cooperative coverage method based on adaptive zoning is characterized by comprising the following steps:
the method comprises the steps that firstly, an optimization problem model for optimizing a target point distribution scheme and a target point traversal sequence of each intelligent agent to maximize detection yield is established for a multi-intelligent-agent heterogeneous target with heterogeneous detection weight values and coverage time; inputting a problem parameter and a genetic algorithm parameter into the model;
secondly, outputting the angle value of each clustering center by adopting a k-means clustering algorithm based on the angular density of the target to complete the partition of the target; wherein the angular density of the target node i is as large as
Wherein theta isiAnd thetajRespectively representing the angles between the connecting lines of the target node I, the target node j and the base station and a horizontal line, wherein delta theta is a set angle threshold value, and I represents a counting function; i is 1,2,3 … m, j is 1,2,3 … m, m is the target point, and the total number i is not equal to j;
thirdly, optimizing the target point traversal sequence in each partition based on an improved genetic algorithm to obtain the optimal traversal sequence of the target points in each partition; the improved genetic algorithm comprises the following steps:
step 1: initializing a population by adopting a hybrid heuristic rule and a Kerries-based heuristic algorithm;
step 2: performing cross operation in an individual cross mode combining the global guidance idea of the particle swarm algorithm;
and step 3: performing mutation operation;
and 4, step 4: calculating the fitness value of individuals in the population;
and 5: updating the population by adopting an elite strategy, which specifically comprises the following steps: sorting all the individuals from big to small based on the objective function value, and selecting the top NpThe good individual enters the next iteration, where NpIs a set value;
step 6: judging whether the end condition of the improved genetic algorithm is reached, if so, continuing the step 7, and if not, returning to the step 2;
and 7: outputting the optimal traversal sequence of the target points in the partitions;
fourthly, calculating the difference value between the execution time of each agent task and the mean value of the execution times of all the agent tasks, and adjusting the corresponding feedback parameters of the zones with the difference values not being 0 according to the difference values;
fifthly, judging whether the current difference value meets the set finishing condition, if not, returning to the second step according to the adjusted feedback parameter, and re-clustering the target point; if yes, continuing the sixth step;
deleting target points which do not meet the time constraint under the obtained optimal traversal sequence, and repairing the optimal traversal sequence;
and seventhly, outputting the traversal sequence of the target points of the repaired intelligent agents.
2. The method of claim 1, wherein in the second step, the method for obtaining the initial center point in the k-means clustering algorithm comprises: selecting the angle of the target with the maximum angular density value as a central point, then removing the selected target point and all points in the angle neighborhood of the selected target point, calculating the angular density of the rest target points, selecting the angle of the target with the maximum angular density value as the central point again, and repeating the process until the number of the central points is the same as that of the intelligent agents;
the method for adjusting the clustering center line comprises the following steps: firstly, obtaining the number of a central line closest to each target point according to a formula (1.12); then, calculating the angle mean value of target points belonging to various types according to the formula (1.13), and updating the angle of the clustering center line; finally, calculating the variation of the central line before and after updating according to the formula (1.14), and if the variation of the central line before and after updating is less than a preset threshold value deltaangleEnding the loop and outputting the angle value of each clustering center, wherein:
in the formula (1.12), σiIndicating the angle value, mu, between the target point i and the base station connecting line and horizontal linejRepresents the angle of the centerline j; in formula (1.13) < mu'jRepresents the angle of the updated centerline j; in the formula (1.14), K represents the number of cluster centers.
3. The method according to claim 1 or 2, wherein in the third step, step 2 of the improved genetic algorithm is implemented in a specific manner as follows:
firstly, a global edge relation table is constructed, and the percentage of the next position of each target point in the existing population, which is connected with target points of different types, is counted in the table; and then crossing the individuals meeting the requirements, randomly selecting chromosome segments on the parent individuals, transforming the parent chromosomes based on the edge relation table, and storing the generated child chromosomes in the population.
4. The method of claim 3, wherein the interleaving operation comprises four interleaving modes, which are as follows:
Cross-I operation: combining the residual genes and the global edge relation table to replace the genes in gs;
Cross-II operation: keeping the gene sequence in gs unchanged, and replacing the residual genes by combining the global edge relation table;
Cross-III operation: keeping the gene sequence in gs unchanged, using the gene sequence as an initial gene segment of a filial generation, and sequentially generating subsequent genes by combining a global edge relation table;
Cross-IV operation: keeping the gene sequence in gs unchanged, using the sequence as the end gene segment of the filial generation, and combining the rest genes with the global edge relation table to generate in sequence.
5. The method of claim 1,2 or 4, wherein in said third step, said step 3 of said improved genetic algorithm comprises mutating said subject by gene point swapping.
6. The method of claim 1,2 or 4, wherein target traversal orders that do not satisfy temporal constraints are repaired using a heuristic rule path repair method.
7. The method of claim 6, wherein the heuristic rules comprise a time-weight-ratio-based heuristic repair rule and a distance-greedy-based heuristic repair rule.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111229814.9A CN114169216A (en) | 2021-10-22 | 2021-10-22 | Multi-agent heterogeneous target cooperative coverage method based on self-adaptive partitioning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111229814.9A CN114169216A (en) | 2021-10-22 | 2021-10-22 | Multi-agent heterogeneous target cooperative coverage method based on self-adaptive partitioning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114169216A true CN114169216A (en) | 2022-03-11 |
Family
ID=80477282
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111229814.9A Pending CN114169216A (en) | 2021-10-22 | 2021-10-22 | Multi-agent heterogeneous target cooperative coverage method based on self-adaptive partitioning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114169216A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060229817A1 (en) * | 2005-04-07 | 2006-10-12 | Jacobson David L | Concurrent two-phase completion genetic algorithm system and methods |
| CN106649456A (en) * | 2016-09-23 | 2017-05-10 | 西安电子科技大学 | Cluster and outlier detection method based on multi-agent evolution |
| CN108880663A (en) * | 2018-07-20 | 2018-11-23 | 大连大学 | Incorporate network resource allocation method based on improved adaptive GA-IAGA |
| CN110163502A (en) * | 2019-05-21 | 2019-08-23 | 北京壹氢科技有限公司 | A kind of more bullet collaboration multistage target assignment methods |
| WO2020000399A1 (en) * | 2018-06-29 | 2020-01-02 | 东莞理工学院 | Multi-agent deep reinforcement learning proxy method based on intelligent grid |
| CN113051807A (en) * | 2021-04-17 | 2021-06-29 | 西北工业大学 | Multi-agent coverage method for complex area based on k-Means and sliding mode variable structure control |
-
2021
- 2021-10-22 CN CN202111229814.9A patent/CN114169216A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060229817A1 (en) * | 2005-04-07 | 2006-10-12 | Jacobson David L | Concurrent two-phase completion genetic algorithm system and methods |
| CN106649456A (en) * | 2016-09-23 | 2017-05-10 | 西安电子科技大学 | Cluster and outlier detection method based on multi-agent evolution |
| WO2020000399A1 (en) * | 2018-06-29 | 2020-01-02 | 东莞理工学院 | Multi-agent deep reinforcement learning proxy method based on intelligent grid |
| CN108880663A (en) * | 2018-07-20 | 2018-11-23 | 大连大学 | Incorporate network resource allocation method based on improved adaptive GA-IAGA |
| CN110163502A (en) * | 2019-05-21 | 2019-08-23 | 北京壹氢科技有限公司 | A kind of more bullet collaboration multistage target assignment methods |
| CN113051807A (en) * | 2021-04-17 | 2021-06-29 | 西北工业大学 | Multi-agent coverage method for complex area based on k-Means and sliding mode variable structure control |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107807665B (en) | Unmanned aerial vehicle formation detection task cooperative allocation method and device | |
| CN110084512B (en) | Multi-robot task allocation method for intelligent warehousing system | |
| Shim et al. | A hybrid estimation of distribution algorithm with decomposition for solving the multiobjective multiple traveling salesman problem | |
| Tan et al. | A hybrid multiobjective evolutionary algorithm for solving vehicle routing problem with time windows | |
| CN107977743B (en) | Multi-unmanned aerial vehicle cooperative task allocation method and device | |
| CN107886201B (en) | Multi-objective optimization method and device for multi-unmanned aerial vehicle task allocation | |
| Dou et al. | Genetic scheduling and reinforcement learning in multirobot systems for intelligent warehouses | |
| CN112013829B (en) | Multi-UAV/UGV collaborative long-term operation path planning method based on multi-objective optimization | |
| CN109544998B (en) | Flight time slot allocation multi-objective optimization method based on distribution estimation algorithm | |
| Luo et al. | Multi-phase modified shuffled frog leaping algorithm with extremal optimization for the MDVRP and the MDVRPTW | |
| CN112327923A (en) | Multi-unmanned aerial vehicle collaborative path planning method | |
| CN110322066B (en) | Collaborative vehicle path optimization method based on shared carrier and shared warehouse | |
| US20220292413A1 (en) | Bike sharing rebalancing optimization method based on adaptive neighborhood search algorithm | |
| CN112084704A (en) | Multi-process route and layout combined optimization method for flexible workshop | |
| CN114399043A (en) | Vehicle position routing method based on non-dominated sorting particle swarm genetic algorithm | |
| CN112801361A (en) | UAVs and UGVs long-term multi-target path planning problem and solving algorithm | |
| CN113867358A (en) | Intelligent path planning method for multi-unmanned vehicle collaborative traversal task | |
| Subbaiah et al. | Scheduling of AGVs and machines in FMS with makespan criteria using sheep flock heredity algorithm | |
| CN113887782A (en) | Genetic-firework mixing method and system for maintenance resource distribution scheduling | |
| CN114169216A (en) | Multi-agent heterogeneous target cooperative coverage method based on self-adaptive partitioning | |
| Mutingi | System reliability optimization: A fuzzy multi-objective genetic algorithm approach | |
| Elsayed et al. | Memetic multi-topology particle swarm optimizer for constrained optimization | |
| Balicki et al. | Multicriteria evolutionary algorithm with tabu search for task assignment | |
| Sabattini et al. | Hierarchical coordination strategy for multi-AGV systems based on dynamic geodesic environment partitioning | |
| CN114415690A (en) | Multi-agent path planning method, navigation server and readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |